[ExI] Trump Is Obsessed With Oil, but Chinese Batteries Will Soon Run the World

[spike at rainier66.com]

-----Original Message-----
From: Keith Henson <hkeithhenson at gmail.com> 
Subject: Re: [ExI] Trump Is Obsessed With Oil, but Chinese Batteries Will Soon Run the World

On Fri, Jan 23, 2026 at 6:50 PM <spike at rainier66.com> wrote:
>
> -----Original Message-----
> From: Keith Henson <hkeithhenson at gmail.com>
...
>
>>.... Coal plants are not at all complicated.

>....The ones I know anything about are.

https://en.wikipedia.org/wiki/Mohave_Power_Station

>...Was a 1580 megawatt electric (MWe) coal-fired power plant that was located in Laughlin, Nevada. Southern California Edison...

Ja, but of course that is for a different application, a much bigger plant than I had imagined, for a different purpose: creating power for a city.

Instead of that big plant, imagine creating power for use by a bunch of microprocessors, with steady use, not much variability in power demand, since they are using the power in reading source material and creating AIs, which they do around the clock.  The human-support stuff doesn't use that power, rather it uses power off the grid.  This power plant imports power for its human staff from distant standard power plants.  Weird, ja?  This plant generates a lot of electric power, but none of it is dumped onto the grid, for it is DC and low voltage, made for consumption by microprocessors.  

This power plant and GPU farm devours fuel, then puts out a lot of heat and AI models.

To estimate the size of the power generator, I am assuming a high end Nvidia GPU, such as an H200, burns up about a kilowatt and costs about $30k.  Sound about right?  So to put together a processor farm, a diesel generator from a retired locomotive could  perhaps run about 2000 of those, with the rest of the power used to (somehow) carry off the waste heat (how?  (convection only?  (Keith, others, ideas please?))) from about 60 million bucks of Nvidia GPUs.

As I think about this, the power generation isn't the real challenge.  Hell that's the easy part.  The real technical challenge is how to carry away the waste heat, how to get sufficient bandwidth for the GPUs to do their research, and how to gather enough investment capital to make it all happen.

spike